Research On Human-Robot Interaction Based On Biosignals

In recent years,robotics technology has developed rapidly and become one of the hot fields that researchers pay attention to.As the core part of robot technology,human-computer interaction(HRI)has been widely used in many fields such as family service,disaster rescue,medical surgery and so on.With the increasing application requirements of HRI,traditional HRI methods(such as cameras etc.)have been unable to meet the task requirements.For the sake of efficiency and safety,researchers have proposed different types of biosignals for the design of HRI systems.The main research content of this paper is to develop a HRI system for mobile robot control based on electroencephalogram(EEG)and electromyogram(EMG).In this paper,the brain-computer interface(BCI)systems based on P300 paradigm and SSVEP paradigm are studied respectively.According to the P300 paradigm,the P300 signals corresponding to different brain regions were analyzed in this paper.And to take it a step further,the electrode configuration of P300 signal acquisition was studied.Based on machine learning model to detect the P300 signal often needs a lot of samples.However,considering the process of EEG signal acquisition is time-consuming.This paper proposes an transfer learning algorithm,Imp Tr Ada Boost,based on Tr Ada Boost.Imp Tr Ada Boost method will train the classification model for the target task with data from source domain as auxiliary samples.Results show that the P300 signal recognition rate has been improved effectively.For SSVEP paradigm,this paper designs experiments to compare the performance of frequency detection algorithm CCA and FBCCA,and the results prove the high efficiency of FBCCA.On this basis,according to the frequency recognition accuracy under different signal processing window length,the optimal window length for the design of BCI system is obtained.In addition,a gesture recognition system based on s EMG is designed in this paper.Firstly,the feature selection algorithm of EMG signal based on GBDT is studied to reduce the dimension of the feature,which solves the over-fitting problem caused by the high complexity of the model.Secondly,the deep learning algorithm of gesture recognition is studied.Considering the advanced characteristics of s EMG signals,the model used for classification needs to have memory function.In this paper,a deep network model combining long and short term memory(LSTM)network and depthwise separable convolution structure is constructed,which effectively improves the online stability of gesture recognition.This study developed a HRI system based on biosignals with the mobile robot ’SAVVY’and verified the effectiveness of proposed algorithms through experiments.